The real cost of AI tools is becoming visible
Uber's $1,500 per month per seat cap on AI tools is getting a lot of attention because it functions as a benchmark. The thread treats it as a useful pricing signal for the industry: if Uber, a sophisticated tech company with genuine engineering needs, draws the line there, that tells you something about what AI tools are actually worth to a typical enterprise user.
The key pushback in the thread is that current token prices are subsidized. The real question is whether $1,500 per month of value holds when prices normalize. One commenter does the math and notes that $18,000 per seat per year starts to make local inference on a $5,000 to $8,000 machine look competitive, especially as 128GB machines capable of running capable local models become available.
This connects to the Claude containment thread, where Anthropic is thinking carefully about what an AI agent can and cannot do. The cost of AI is not just the token bill. It includes the operational risk of what these systems do when unsupervised.
So what?
If you are building AI tooling for enterprises, Uber's cap is your de facto ceiling for per-seat pricing at scale. Price above $1,500 per month and you will face procurement resistance. The smarter play may be outcome-based pricing that sidesteps the token cost conversation entirely.